update aggregate_metric arg

lm_eval/tasks/benchmarks/flan/flan_held_in.yaml

@@ -4,8 +4,9 @@ task:
   # ANLI R1
   - group: anli_r1_flan
     group_alias: ANLI R1
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: anli_r1
         task_alias: prompt-0
@@ -55,8 +56,9 @@ task:
   # ANLI R2
   - group: anli_r2_flan
     group_alias: ANLI R2
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: anli_r2
         task_alias: prompt-0
@@ -106,8 +108,9 @@ task:
   # ANLI R3
   - group: anli_r3_flan
     group_alias: ANLI R3
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: anli_r3
         task_alias: prompt-0
@@ -157,8 +160,9 @@ task:
   # Arc Easy
   - group: arc_easy_flan
     group_alias: Arc Easy
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: arc_easy
         task_alias: prompt-0
@@ -198,8 +202,9 @@ task:
   # Arc Challenge
   - group: arc_challenge_flan
     group_alias: Arc Challenge
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: arc_challenge
         task_alias: prompt-0
@@ -239,8 +244,9 @@ task:
   # BoolQ
   - group: boolq_flan
     group_alias: BoolQ
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: boolq
         task_alias: prompt-0
@@ -295,8 +301,9 @@ task:
   # RTE
   - group: rte_flan
     group_alias: RTE
-    group_config:
-      aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
     task:
       - task: rte
         task_alias: prompt-0

lm_eval/tasks/mmlu/continuation/_mmlu.yaml

@@ -4,24 +4,29 @@ task:
   - group: stem
     task:
       - mmlu_continuation_stem
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: other
     task:
       - mmlu_continuation_other
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: social sciences
     task:
       - mmlu_continuation_social_sciences
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: humanities
     task:
       - mmlu_continuation_humanities
-    aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric:
+  - metric: acc
     weight_by_size: True
-aggregate_metric: True
-weight_by_size: True
 metadata:
   version: 1

lm_eval/tasks/mmlu/default/_mmlu.yaml

@@ -4,7 +4,8 @@ task:
   - mmlu_other
   - mmlu_social_sciences
   - mmlu_humanities
-aggregate_metric: True
-weight_by_size: True
+aggregate_metric:
+  - metric: acc
+    weight_by_size: True
 metadata:
   version: 1

lm_eval/tasks/mmlu/default/_mmlu_humanities.yaml

@@ -2,6 +2,8 @@ group: mmlu_humanities
 group_alias: humanities
 task:
   - mmlu_humanities_tasks
-aggregate_metric: True
-weight_by_size: True
-version: 1
+aggregate_metric:
+  - metric: acc
+    weight_by_size: True
+metadata:
+  version: 1

lm_eval/tasks/mmlu/default/_mmlu_other.yaml

@@ -2,6 +2,8 @@ group: mmlu_other
 group_alias: other
 task:
   - mmlu_other_tasks
-aggregate_metric: True
-weight_by_size: True
-version: 1
+aggregate_metric:
+  - metric: acc
+    weight_by_size: True
+metadata:
+  version: 1
\ No newline at end of file

lm_eval/tasks/mmlu/default/_mmlu_social_sciences.yaml

@@ -2,6 +2,8 @@ group: mmlu_social_sciences
 group_alias: social sciences
 task:
   - mmlu_social_sciences_tasks
-aggregate_metric: True
-weight_by_size: True
-version: 1
+aggregate_metric:
+  - metric: acc
+    weight_by_size: True
+metadata:
+  version: 1
\ No newline at end of file

lm_eval/tasks/mmlu/default/_mmlu_stem.yaml

@@ -2,6 +2,8 @@ group: mmlu_stem
 group_alias: stem
 task:
   - mmlu_stem_tasks
-aggregate_metric: True
-weight_by_size: True
-version: 1
+aggregate_metric:
+  - metric: acc
+    weight_by_size: True
+metadata:
+  version: 1
\ No newline at end of file

lm_eval/tasks/mmlu/flan_cot_fewshot/_mmlu.yaml

@@ -4,24 +4,29 @@ task:
   - group: stem
     task:
       - mmlu_flan_cot_fewshot_stem
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: other
     task:
       - mmlu_flan_cot_fewshot_other
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_cot_fewshot_social_sciences
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_cot_fewshot_humanities
-    aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric:
+  - metric: acc
     weight_by_size: True
-aggregate_metric: True
-weight_by_size: True
 metadata:
   version: 1

lm_eval/tasks/mmlu/flan_cot_zeroshot/_mmlu.yaml

@@ -4,24 +4,29 @@ task:
   - group: stem
     task:
       - mmlu_flan_cot_zeroshot_stem
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: other
     task:
       - mmlu_flan_cot_zeroshot_other
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_cot_zeroshot_social_sciences
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_cot_zeroshot_humanities
-    aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric:
+  - metric: acc
     weight_by_size: True
-aggregate_metric: True
-weight_by_size: True
 metadata:
   version: 1

lm_eval/tasks/mmlu/flan_n_shot/generative/_mmlu.yaml

@@ -4,24 +4,29 @@ task:
   - group: stem
     task:
       - mmlu_flan_n_shot_generative_stem
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: other
     task:
       - mmlu_flan_n_shot_generative_other
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_n_shot_generative_social_sciences
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_n_shot_generative_humanities
-    aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric:
+  - metric: acc
     weight_by_size: True
-aggregate_metric: True
-weight_by_size: True
 metadata:
   version: 1

lm_eval/tasks/mmlu/flan_n_shot/loglikelihood/_mmlu.yaml

@@ -4,24 +4,29 @@ task:
   - group: stem
     task:
       - mmlu_flan_n_shot_loglikelihood_stem
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: other
     task:
       - mmlu_flan_n_shot_loglikelihood_other
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: social sciences
     task:
       - mmlu_flan_n_shot_loglikelihood_social_sciences
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: humanities
     task:
       - mmlu_flan_n_shot_loglikelihood_humanities
-    aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric:
+  - metric: acc
     weight_by_size: True
-aggregate_metric: True
-weight_by_size: True
 metadata:
   version: 1

lm_eval/tasks/mmlu/generative/_mmlu.yaml

@@ -4,24 +4,29 @@ task:
   - group: stem
     task:
       - mmlu_stem_generative
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: other
     task:
       - mmlu_other_generative
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: social sciences
     task:
       - mmlu_social_sciences_generative
-    aggregate_metric: True
-    weight_by_size: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
   - group: humanities
     task:
       - mmlu_humanities_generative
-    aggregate_metric: True
+    aggregate_metric:
+      - metric: acc
+        weight_by_size: True
+aggregate_metric:
+  - metric: acc
     weight_by_size: True
-aggregate_metric: True
-weight_by_size: True
 metadata:
   version: 1